Method, system and device for detecting positive lymph nodes in medical image and medium

Abstract

The invention discloses a method, system and device for detecting positive lymph nodes in a medical image and a medium, and relates to the technical field of medical image detection.The method comprises the steps that a preoperative CT image of a sample individual is collected; inputting the preoperative CT image into the image detection model, and extracting positive lymph node image features of different areas in the preoperative CT image through a multi-scale feature extraction backbone network containing P1, P2, P3, P4 and P5 layers; fusing the extracted multi-scale features by adopting a multi-scale feature fusion module, and identifying the regional position of the positive lymph node in the medical image according to the fused features; the method shows superiority in the aspects of automatic positioning and detection of the positive lymph nodes of different organ parts in the medical image, and the detection precision of the multi-organ part and multi-target positive lymph nodes is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image detection, and in particular to a method, system, equipment and medium for detecting positive lymph nodes in medical images. Background Art

[0002] Imaging examination is the main method for diagnosing lymph node metastasis (LNM). The determination of the scope of surgery usually depends on the imaging diagnosis. Therefore, accurate lymph node detection and avoiding missing LNM are crucial for determining the lymph node radiotherapy target.

[0003] At present, lymph node detection mainly relies on clinicians to make diagnoses based on imaging features. For example, some scholars have proposed a radiomics strategy based on preoperative MRI tumor area for binary prediction of bladder cancer lymph node metastasis (WuS, et al, Development and Validation of an MRI-Based Radiomics Signature for the Preoperative Prediction of Lymph Node Metastasis in Bladder Cancer, EBioMedicine, 2018Aug: 34: 76-84). This study extracted imaging features from the tumor area of ​​preoperative MRI images, and after feature screening and logistic regression, constructed a quantitative prediction model - nomogram, which was used to determine whether lymph node metastasis occurred in patient samples.

[0004] However, the current methods based on intelligent image analysis are still at the level of binary classification of whether the tumor has lymph node metastasis, and have not yet gone deep into the localization detection of tumor lymph node metastasis. The detection methods used are difficult to further locate the positive lymph node locations in different parts of the body in preoperative images. Summary of the invention

[0005] In view of the deficiency that the prior art cannot accurately locate positive lymph nodes, the present invention proposes a method, system, device and medium for detecting positive lymph nodes in medical images, and solves the problems existing in the prior art by constructing a detection framework for positive lymph nodes in different regional positions, postures and shapes.

[0006] A method for detecting positive lymph nodes in a medical image comprises the following steps:

[0007] Collect preoperative CT images of the sample individuals;

[0008] The preoperative CT image is input into the image detection model, and the positive lymph node image features of different regions in the preoperative CT image are extracted through the multi-scale feature extraction backbone network including P1, P2, P3, P4, and P5 layers. Specifically, the preoperative CT image is input into the P1 layer and the P2 layer for convolution operation, and then input into the bidirectional decomposition VSS module, and respectively passes through different linear layers and their convolution layers to generate forward features and backward features, and the forward features and backward features are spliced ​​to obtain output features. The output features of the P2 layer are operated in the P3 layer and then input into the deformable convolution module DCNv3 to extract the dominant features of positive lymph nodes in different regions. The dominant features of positive lymph nodes in different regions extracted from the P3 layer are operated in the P4 layer and the P5 layer to obtain the positive lymph node image features in different regions; the positive lymph node image features in different regions are fused by the multi-scale feature fusion module to obtain the fused features;

[0009] The region containing positive lymph node images is identified based on the fused features.

[0010] Furthermore, the P1, P2, P3, P4, and P5 layers all include a convolution operation, a batch normalization operation, and an activation function; the activation function adopts the LeakyHS activation function, which is expressed as:

[0011]

[0012] Among them, x represents the input feature.

[0013] Furthermore, the preoperative CT image is input into the P1 layer and the P2 layer for convolution operation, and then input into the bidirectional decomposition VSS module, and passes through different linear layers and convolution layers respectively to generate forward features and backward features, and the forward features and the backward features are spliced ​​to obtain output features, including the following steps:

[0014] Given a set of feature sequences x, after inputting them into the bidirectional decomposition VSS module, they pass through different linear layers and convolutional layers to generate forward features x f With the backward feature x b , respectively expressed as:

[0015]

[0016] Among them, ForwardConv represents the forward convolution operation, and BackwardConv represents the deconvolution operation;

[0017] Using the forward and backward state space model SSM to f and x b Perform forward and backward scan calculations respectively, and use the same weight to obtain the new forward feature x'f and the backward feature x' b ;

[0018]

[0019] The forward and backward feature gates are constructed based on the linear layer and the activation function LeakyHS, and x' f and x' b After element-by-element point multiplication, concatenation is performed to obtain the output feature x':

[0020] x'=Concat(x' f *LeakyHS(Linear(x)),x' b *LeakyHS(Linear(x)).

[0021] Furthermore, the positive lymph node dominant features of different regions extracted by the P3 layer are calculated in the P4 layer and the P5 layer to obtain the positive lymph node image features of different regions, which specifically includes the following steps:

[0022] The dominant features of positive lymph nodes in different regions extracted by the p3 layer are input into the C2f module for splicing and compression after being calculated in the P4 layer;

[0023] The concatenated and compressed features are input into the C2f and SPFF modules after the P5 layer operation to extract the positive lymph node image features in different regions.

[0024] Furthermore, the multi-scale feature fusion module is used to fuse the extracted multi-scale features, which specifically includes the following steps:

[0025] The upsampling operation is used to integrate the large-scale feature information of the P4 and P5 layers into the p3 layer;

[0026] The feature information of different scales is further fused through the cross-stage partial bottleneck layer C2f with two-way convolution; at the same time, the GAM module combining dual channels and spatial attention is used to extract global attention to obtain the fused features.

[0027] It also includes using a loss function to discriminate the region containing the positive lymph node image; the loss function includes a classification loss VFL, a regression loss DFL and an IoUL, which are respectively expressed as:

[0028]

[0029] Among them, α and γ are hyperparameters, p γ is the adjustment factor, the value of q is the IOU between the predicted box and the true box, and p represents the classification probability;

[0030]

[0031] Among them, B gt The rectangular box indicating the image location of the positive lymph node, B predict Represents the predicted rectangular box; Intersection represents the overlapping area between the real target box and the predicted target box, and Union represents the combined area of ​​the two;

[0032] DFL(S i ,S i+1 )=-((y i+1 -y)log(S i )+(yy i )log(S i+1 ))

[0033] Among them, y i represents the bounding box value predicted for the i-th time; y is the actual bounding box value;

[0034] The total loss function Total Loss is expressed as:

[0035]

[0036] Among them, n pos is the number of feature points assigned as positive samples, is the balance coefficient of the regression loss.

[0037] The present invention also includes a system for detecting positive lymph nodes in a medical image, comprising:

[0038] An acquisition module, used to acquire preoperative CT images of individual samples;

[0039] The feature extraction module is used to input the preoperative CT image into the image detection model, and extract the positive lymph node image features of different regions in the preoperative CT image through a multi-scale feature extraction backbone network including P1, P2, P3, P4, and P5 layers. Specifically, the preoperative CT image is input into the P1 layer and the P2 layer for convolution operation, and then input into the bidirectional decomposition VSS module, and respectively passes through different linear layers and their convolution layers to generate forward features and backward features, and the forward features and backward features are spliced ​​to obtain output features. The output features of the P2 layer are input into the deformation convolution module DCNv3 after being operated in the P3 layer to extract the dominant features of positive lymph nodes in different regions, and the dominant features of positive lymph nodes in different regions extracted from the P3 layer are operated in the P4 layer and the P5 layer to obtain the positive lymph node image features of different regions; the multi-scale feature fusion module is used to fuse the positive lymph node image features of different regions to obtain the fused features;

[0040] The recognition module is used to identify the area containing positive lymph node images according to the fused features.

[0041] The present invention also includes a computer device for detecting positive lymph nodes in medical images, including: a memory, a processor, and a computer program stored in the memory, and when the processor executes the computer program, the steps of the method for detecting positive lymph nodes in medical images are implemented.

[0042] The present invention also includes a readable storage medium, which stores a computer program. The computer program includes program instructions. When the program instructions are executed by a processor, they are used to execute the steps of the method for detecting positive lymph nodes in medical images.

[0043] The present invention provides a method, system, device and medium for detecting positive lymph nodes in medical images, which have the following beneficial effects:

[0044] The present invention designs a multi-scale feature extraction backbone network, adopts a five-layer feature pyramid structure, and innovatively introduces a VSS module and a DCNv3 module in the corresponding layer. The VSS module solves the problem that it is difficult to extract the tiny target features of positive lymph nodes contained in the large background of CT images. The DCNv3 module enhances the network's recognition ability of positive lymph nodes of different body parts, regional positions, shapes and postures. The two modules are combined to effectively extract semantic features, and multi-regional, tiny, and complex-shaped and postured positive lymph nodes are represented. Finally, a multi-scale feature fusion module is used to fuse the extracted multi-scale features to identify the area containing the positive lymph node image. The method constructs a detection framework for positive lymph nodes of different body parts, regional positions, and posture shapes, and is superior in automatically locating, detecting, and visually presenting positive lymph nodes in different parts of the patient, and can improve the prediction accuracy of multi-part and multi-regional lymph nodes. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a flowchart of a method for detecting positive lymph nodes in a medical image in an embodiment of the present invention;

[0046] Figure 2 Schematic diagram of a method for detecting positive lymph nodes in a medical image in an embodiment of the present invention;

[0047] Figure 3 This is a schematic diagram of the VSS module architecture in an embodiment of the present invention;

[0048] Figure 4 Schematic diagram of the GAM module architecture in an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0050] The present invention proposes a multi-region positive lymph node detection technology. According to the biological characteristics of multi-region lymph nodes and their dominant signs in preoperative enhanced CT, a multi-scale deformable convolution feature extraction module is designed, which contains a multi-scale visual state space submodule and a deformable convolution network module to fully explore the important features of multiple, tiny, distorted and deformed targets (positive lymph nodes) hidden in the image area. On this basis, a feature / attention fusion module is designed, including multi-scale feature fusion and global attention fusion mechanisms, to fully integrate feature information of different scales and network attention. Finally, a detection output module is designed to perform detection output at three resolution levels and select the optimal detection result. Figure 1 As shown, the detection method specifically comprises the following steps:

[0051] S1. Preoperative enhanced CT images of patient samples.

[0052] S2, positive lymph node feature extraction; a multi-scale feature extraction backbone network was designed, and a feature pyramid with five layers of P1, P2, P3, P4, and P5 was constructed, such as Figure 2 As shown. P1 to P5 all contain convolution, batch normalization, activation function and other operation units. Among them, the innovative LeakyHardSwish (Leaky HS) activation function is designed, as shown in formula (1), to effectively solve the problem of network gradient disappearance and enhance the convergence speed and model generalization performance. After the P2 layer, a bidirectional decomposition multi-scale visual state space (Visual State Space, VSS) submodule is designed with Gate operation to fully solve the problem of extracting the features of small targets of positive lymph nodes contained in the large background of CT images.

[0053]

[0054] The features output from the P1 operation unit enter the next scale operation unit P2, which also contains convolution, batch normalization, and activation function LeakyHS to further extract the high-dimensional information of the image. A VSS module is designed behind the P2 operation unit, such as Figure 3 Given a set of feature sequences x, when it is input into the bidirectional decomposition VSS module, it passes through different linear layers and convolutional layers to generate forward features x f With the backward feature x b, as shown in formula (2). Then, the forward / backward state space model (SSM) is used to calculate the above x f and x b Perform forward and backward scan calculations separately and share weights to obtain new forward features x' f and the backward feature x' b , as shown in formula (3). Finally, the forward and backward feature gates are constructed based on the linear layer and the activation function LeakyHS, and are combined with x' f and x' b The element-by-element point multiplication and concatenation are performed to form the output feature x', as shown in formula (4). The purpose of using the Gate operation is to reduce the directional deviation introduced during forward / backward scanning and enhance the connection characteristics of different image regions.

[0055]

[0056] x'=Concat(x' f *LeakyHS(Linear(x)),x' b *LeakyHS(Linear(x)) (4)

[0057] The P3, P4, and P5 layers are designed for higher-scale feature extraction, and their layer structures are similar to those of the P1 and P2 layers. The Deformable Convolutional Networks v3 (DC Nv3) module is innovatively introduced after the P3 layer to fully explore the important features of multiple, small, and distorted positive lymph nodes hidden in the image area, such as Figure 2 As shown in the figure, it aims to enhance the network's ability to recognize positive lymph nodes in different body parts, regional locations, shapes and postures. DCNv3 is a variable convolution module that draws on the idea of ​​separable convolution and converts the original convolution weight w g It is separated into a depth part and a point-by-point part, where the depth part is composed of the original position-aware modulation scalar m gk The point-by-point part is the projection weight w shared between the sampling points, which aims to enhance the network's recognition ability of positive lymph nodes in different body parts, regional locations, shapes and postures. The specific expression is shown in formula (5).

[0058]

[0059] Where p0 is any element of the input feature matrix; G represents the total number of aggregation groups, and w g represents the position-independent projection weight of the group; represents the modulation scalar of the k-th sampling point in the g-th group, normalized by the softmax function along dimension K; xg represents the input feature matrix of the gth group; Δp gk is the grid sampling position p in group g k The corresponding offset. The C2f module is introduced after the P4 layer. The C2f module has two functions: one is feature aggregation, which effectively extracts and fuses multi-scale feature information by splicing the features output by different Bottleneck modules and the original features input by the P4 layer, and realizes effective representation of the detection target; the other is model compression. After the multi-scale feature fusion, C2f uses convolution operation to effectively compress the feature map, realize the lightweight model and effectively maintain the expressive ability of the features. The C2f and SPFF modules are introduced at the same time after the P5 layer to further realize multi-scale feature extraction and model lightweight.

[0060] S3, multi-scale feature / global attention fusion; In order to fully integrate the different scale feature information output by the P3, P4, and P5 layers, the large-scale feature information of the P4 and P5 layers is first integrated into the P3 layer using an upsampling operation, and then the feature information of different scales is further fused through the cross-stage partial bottleneck layer with two convolutions (Cross Stage Partial Bottom Neck wit h 2Convolutions, C2f).

[0061] Since the captured low-level semantic features may not be sufficient to detect and distinguish the positive lymph node micro-targets, the present invention designs a multi-scale feature / global attention fusion module before detection, and uses upsampling with C2f to effectively fuse the information of the high-order semantic feature P5 layer and the low-order semantic features P3 and P4. On this basis, an innovative GAM module with dual channels and spatial attention integration is designed, such as Figure 4 As shown, the global attention is fully extracted to better understand the relationship between the detection target and the global structure of the image, realize multi-region and small target perception, maximize the fusion of channel and spatial attention, and realize the effective characterization of positive lymph nodes.

[0062] S4, multi-resolution detection output; the present invention corresponds to the P3, P4, and P5 layers, and designs corresponding detection outputs respectively, so that the network can adapt to tiny targets of different scales and resolutions, improve the perception ability of multi-scale targets, and improve the detection accuracy of positive lymph nodes.

[0063] The loss function consists of two branches: classification loss Varifocal Loss (VFL), and regression loss Distribution Focal Loss (DFL) and Intersection of Union Loss (IoUL), which correspond to positive lymph node discrimination and accurate detection, respectively.

[0064]

[0065] Among them, α and γ are hyperparameters. α is a balance parameter used to adjust the weights of positive and negative samples, and the adjustment factor p γ It can reduce the impact of easy-to-classify samples on the loss, making the model pay more attention to the difficult-to-identify positive lymph node samples. For negative samples, when q = 0, then p γ It can be used to reduce the contribution of negative samples to the loss, while for positive samples (i.e., when q>0), the value of q is the IOU between the predicted box and the true box. q is used to weight positive samples, so that when the IOU of positive samples is higher, its contribution to the loss will also increase, so that the model pays more attention to high-quality positive samples, which is conducive to improving the model's detection accuracy for positive lymph nodes.

[0066]

[0067] Among them, B gt and B predict The rectangular boxes represent the locations of positive lymph nodes and the predicted rectangular boxes respectively.

[0068] DFL(S i ,S i+1 )=-((y i+1 -y)log(S i )+(yy i )log(S i+1 )) (8)

[0069] Based on the above loss function, the total loss function Total Loss can be expressed as:

[0070]

[0071] Among them, n pos is the number of feature points assigned as positive samples, is the balance coefficient of the regression loss.

[0072] S5. Visualization of positive lymph node detection results; In order to improve the interpretability of the model and increase the credibility of the model, Grad-CAM is introduced to realize the visualization process of model detection by using heat map, while evaluating the positioning ability of model detection and providing users with a visual explanation of the model.

[0073] The present invention constructs a positive lymph node feature extraction module, which introduces VSS and DCNv3 structures, can identify and memorize the positive lymph node features in different regions, and greatly filter out irrelevant information, thereby improving the network's ability to mine important features of positive lymph nodes. After obtaining the important features of the positive lymph nodes, in order to further visualize the network prediction results and enhance their interpretability, a method of global attention mechanism is introduced. This method can amplify the global dimensional interaction features while reducing information diffusion, allowing the model to perform feature modeling on positive lymph nodes at different spatial scales, so that the network can adapt to targets of different scales and resolutions, and improve the perception of multi-scale targets to ensure the detection accuracy of positive lymph nodes.

[0074] Based on the same inventive concept, the present invention also proposes a detection system for positive lymph nodes in medical images, comprising:

[0075] The acquisition module is used to acquire preoperative CT images of individual samples.

[0076] The feature extraction module is used to input the preoperative CT image into the image detection model, and extract the positive lymph node image features of different regions in the preoperative CT image through a multi-scale feature extraction backbone network including P1, P2, P3, P4, and P5 layers. Specifically, the preoperative CT image is input into the P1 layer and P2 layer for convolution operation, and then input into the bidirectional decomposition VSS module, and passes through different linear layers and their convolution layers respectively to generate forward features and backward features, and the forward features and backward features are concatenated to obtain output features. The output features of the P2 layer are operated in the P3 layer and then input into the deformable convolution module DCNv3 to extract the dominant features of positive lymph nodes in different regions. The dominant features of positive lymph nodes in different regions extracted by the P3 layer are operated in the P4 and P5 layers to obtain the positive lymph node image features of different regions; the multi-scale feature fusion module is used to fuse the positive lymph node image features of different regions to obtain the fused features.

[0077] The recognition module is used to identify the area containing positive lymph node images according to the fused features.

[0078] The present invention also proposes a computer device for detecting positive lymph nodes in medical images, comprising: a memory, a processor, and a computer program stored in the memory, and the processor implements the steps of a method for detecting positive lymph nodes in medical images when executing the computer program.

[0079] The present invention also provides a readable storage medium, which stores a computer program. The computer program includes program instructions. When the program instructions are executed by a processor, they are used to execute the steps of a method for detecting positive lymph nodes in a medical image.

[0080] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for detecting positive lymph nodes in medical images, characterized in that: The following steps are involved: Collect preoperative CT images of the sample individuals; The preoperative CT image is input into the image detection model, and the positive lymph node image features of different regions in the preoperative CT image are extracted through the multi-scale feature extraction backbone network including P1, P2, P3, P4, and P5 layers. Specifically, the preoperative CT image is input into the P1 layer and the P2 layer for convolution operation, and then input into the bidirectional decomposition VSS module, and respectively passes through different linear layers and their convolution layers to generate forward features and backward features, and the forward features and backward features are spliced ​​to obtain output features. The output features of the P2 layer are operated in the P3 layer and then input into the deformable convolution module DCNv3 to extract the dominant features of positive lymph nodes in different regions. The dominant features of positive lymph nodes in different regions extracted from the P3 layer are operated in the P4 layer and the P5 layer to obtain the positive lymph node image features in different regions; the positive lymph node image features in different regions are fused by the multi-scale feature fusion module to obtain the fused features; The region containing positive lymph node images is identified based on the fused features.

2. The method for detecting positive lymph nodes in medical images according to claim 1, characterized in that: The P1, P2, P3, P4, and P5 layers all include convolution operations, batch normalization operations, and activation functions; the activation function uses the LeakyHS activation function, which is expressed as: Among them, x represents the input feature.

3. The method for detecting positive lymph nodes in medical images according to claim 1, characterized in that: The preoperative CT image is input into the P1 layer and the P2 layer for convolution operation, and then input into the bidirectional decomposition VSS module, and passes through different linear layers and convolution layers respectively to generate forward features and backward features, and the forward features and the backward features are spliced ​​to obtain output features, including the following steps: Given a set of feature sequences x, after inputting them into the bidirectional decomposition VSS module, they pass through different linear layers and convolutional layers to generate forward features x f With the backward feature x b , respectively expressed as: Among them, ForwardConv represents the forward convolution operation, and BackwardConv represents the deconvolution operation; Using the forward and backward state space model SSM to f and x b Perform forward and backward scan calculations respectively, and use the same weight to obtain the new forward feature x' f and the backward feature x' b ; The forward and backward feature gates are constructed based on the linear layer and the activation function LeakyHS, and x' f and x' b After element-by-element point multiplication, concatenation is performed to obtain the output feature x': x'=Concat(x' f *LeakkyHS(Lnear(x)),x' b *LeakyHS(Linear(x))。 4. The method for detecting positive lymph nodes in medical images according to claim 1, characterized in that: The method of obtaining positive lymph node image features of different regions after performing operations on the P4 and P5 layers on the dominant features of positive lymph nodes extracted from the P3 layer includes the following steps: The dominant features of positive lymph nodes in different regions extracted by the P3 layer are input into the C2f module for splicing and compression after being calculated in the P4 layer; The concatenated and compressed features are input into the C2f and SPFF modules after the P5 layer operation to extract the positive lymph node image features in different regions.

5. The method for detecting positive lymph nodes in medical images according to claim 1, characterized in that: The multi-scale feature fusion module is used to fuse the extracted multi-scale features, specifically comprising the following steps: The scale feature information of P4 and P5 layers is integrated into P3 layer by upsampling operation; The feature information of different scales is further fused through the cross-stage partial bottleneck layer C2f with two-way convolution; at the same time, the GAM module combining dual channels and spatial attention is used to extract global attention to obtain the fused features.

6. The method for detecting positive lymph nodes in medical images according to claim 1, characterized in that: It also includes using a loss function to discriminate the region containing the positive lymph node image; the loss function includes a classification loss VFL, a regression loss DFL and an IoUL, which are respectively expressed as: Among them, α and γ are hyperparameters, p γ is the adjustment factor, the value of q is the IOU between the predicted box and the real box, and p is the target classification probability; Among them, B gt The rectangular box representing the image location of the positive lymph node, B predict Represents the predicted rectangular box; Intersection represents the overlapping area between the real target box and the predicted target box, and Union represents the combined area of ​​the real target box and the predicted target box; DFL(S i ,S i+1 )=-((y i+1 -y)log(S i )+(y-y i )log(S i+1 )) Among them, y i represents the bounding box value predicted for the i-th time; y is the actual bounding box value; The total loss function Total Loss is expressed as: Among them, n pos is the number of feature points assigned as positive samples, is the balance coefficient of the regression loss.

7. A system for detecting positive lymph nodes in medical images, characterized in that: include: An acquisition module, used to acquire preoperative CT images of individual samples; The feature extraction module is used to input the preoperative CT image into the image detection model, and extract the positive lymph node image features of different regions in the preoperative CT image through a multi-scale feature extraction backbone network including P1, P2, P3, P4, and P5 layers. Specifically, the preoperative CT image is input into the P1 layer and the P2 layer for convolution operation, and then input into the bidirectional decomposition VSS module, and respectively passes through different linear layers and their convolution layers to generate forward features and backward features, and the forward features and backward features are spliced ​​to obtain output features. The output features of the P2 layer are input into the deformation convolution module DCNv3 after being operated in the P3 layer to extract the dominant features of positive lymph nodes in different regions, and the dominant features of positive lymph nodes in different regions extracted from the P3 layer are operated in the P4 layer and the P5 layer to obtain the positive lymph node image features of different regions; the multi-scale feature fusion module is used to fuse the positive lymph node image features of different regions to obtain the fused features; The recognition module is used to identify the area containing positive lymph node images according to the fused features.

8. A computer device for detecting positive lymph nodes in medical images, characterized in that: include: A memory, a processor, and a computer program stored in the memory, wherein the processor implements the steps of the method for detecting positive lymph nodes in a medical image according to any one of claims 1 to 6 when executing the computer program.

9. A readable storage medium, characterized in that: The readable storage medium stores a computer program, which includes program instructions. When the program instructions are executed by a processor, they are used to execute the steps of the method for detecting positive lymph nodes in a medical image according to any one of claims 1 to 6.

Images